mistral.md

Mistral

Auto-instrument Mistral AI API calls with full tracing, token counting, and cost estimation.

Installation

npm install @openlantern-ai/mistral @mistralai/mistralai

@openlantern-ai/mistral has a peer dependency on @openlantern-ai/sdk — install it if you haven't already:

npm install @openlantern-ai/sdk

Setup

import { Mistral } from "@mistralai/mistralai";
import { LanternTracer, LanternExporter } from "@openlantern-ai/sdk";
import { wrapMistralClient } from "@openlantern-ai/mistral";

const tracer = new LanternTracer({
  serviceName: "my-agent",
  exporter: new LanternExporter({
    endpoint: "https://your-lantern-instance.com",
    apiKey: process.env.LANTERN_API_KEY!,
  }),
});

const client = wrapMistralClient(new Mistral({ apiKey: process.env.MISTRAL_API_KEY! }), tracer);

Usage

// All calls are now traced automatically
const response = await client.chat.complete({
  model: "mistral-large-latest",
  messages: [{ role: "user", content: "Explain quantum computing in one paragraph." }],
});

// Tool calls are captured as child spans
const toolResponse = await client.chat.complete({
  model: "mistral-large-latest",
  messages: [{ role: "user", content: "What's the weather in London?" }],
  tools: [{
    type: "function",
    function: {
      name: "get_weather",
      description: "Get weather for a city",
      parameters: {
        type: "object",
        properties: { city: { type: "string" } },
      },
    },
  }],
});

await tracer.shutdown();

What Gets Traced

Span Type	Created When	Data Captured
llm_call	Each chat.complete() call	Input messages, output text, model, prompt/completion tokens, estimated cost, finish reason
tool_call	Each tool call in the response	Tool name, arguments (from function.arguments), parent LLM span

Token counts are read from Mistral's camelCase response fields: response.usage.promptTokens and response.usage.completionTokens. The finish reason comes from choice.finishReason (also camelCase).

Options

Option	Type	Description
traceId	string	Attach spans to an existing trace instead of creating a new one per call
agentName	string	Agent name for auto-created traces (default: "mistral-agent")

Troubleshooting

• Method is chat.complete(), not chat.completions.create(): Mistral's SDK uses a different method name than OpenAI. The wrapper instruments client.chat.complete().
• Token field names are camelCase: Unlike OpenAI (prompt_tokens), Mistral uses promptTokens and completionTokens. The wrapper handles this automatically.
• Streaming not traced: The wrapper instruments the non-streaming chat.complete() path. For streaming, use the LLM Proxy instead.